The most quoted idea in professional services right now is also the most correct, which is why it deserves a careful disagreement rather than a lazy one. The claim goes like this: for a century, consulting sold a knowledge gap — I know something you don’t — and the price of knowing is now collapsing toward zero. So, the value migrates from knowing to doing. From “here is a strategy” to “I opened the market and here are your first ten customers.” From “here is an operating model” to “I rebuilt the org and it is running.” Sell the outcome, not the advice. Become the operator.

It is right. For consulting, it is exactly right, and the firms acting on it will take the ground from the firms that don’t. But it is a statement about one seat at the capital table, and it is being read as if it were a statement about all of them. There is a seat the argument does not touch, because that seat was never selling the knowing in the first place. It runs a panel. It is a lab.

Think about what a medical lab actually sells, because almost everyone gets it wrong. It does not sell knowledge — the reference ranges are printed in every textbook and now sit inside every model. It does not sell doing — it performs no surgery, prescribes no drug, manages no recovery. What it sells is narrower and far more durable than either: a result — reproducible, independent, and trusted by every party who has to act on it. The value was never the knowing and never the doing. It was the specimen, drawn under controlled conditions, run through a defined panel, returned as a number the treating parties can build a decision on without having to trust each other.

Cheap intelligence does nothing to that business. You cannot prompt your way to a blood draw. The specimen has to be collected correctly or everything downstream is noise, and no drop in the cost of inference changes that. If anything, the cheaper analysis gets, the more a lab is worth — because when everyone can generate a confident-sounding read of whatever data they are handed, the scarce thing becomes a result nobody can generate, dial, or argue with. The lab was always immune to the operator thesis. Nobody has ever said the collapse in the cost of knowing means the pathology lab should start doing the operations. The sentence is absurd on its face, and it is absurd for a structural reason worth naming: the moment the lab has a stake in the treatment, its result is no longer trusted, and it has destroyed the only thing it was selling.

This is the seat the read occupies. Not the strategist’s seat, which the operator thesis correctly hollows out. Not the operator’s seat, which belongs — permanently, and by design — to the company and the capital behind it. The read draws a specimen and runs a panel on the one thing traditional diligence cannot collect: what happens inside a leadership team when capital arrives, taken independently, from each seat, so the contradictions surface instead of getting negotiated away in the room. It returns a result the company and its investor can both act on. And then it stops.

And a result is a finding, not an instruction. The physician who orders an advanced diagnostic does not hand over their judgement — they get a reading they could not have produced on their own, and they read it against everything the lab cannot see: the patient’s history, the tolerances, the alternatives. The capital owner holds the same position. A verdict that says Revisit is not a command; it is a finding. The investor who reads it and deploys anyway, because the thesis carries the risk, has not misused the instrument — they have used it exactly as intended: it told them what breaks, and they made the call. The decision was always theirs. The lab just made sure nothing structural went unseen on the way to it.

Which brings me to the part of a lab that almost nobody outside one thinks about, and which turns out to be the whole argument.

A serious lab does not simply run whatever arrives and publish whatever comes out. It validates the specimen first, and it refuses to certify a reading it can see is corrupted at the source. When a panel returns a value physically incompatible with the patient being alive, the lab does not report it. It flags the draw, goes back to the source, and re-runs. That requery is not the lab leaving its scope to go practise medicine. It is the lab running the panel correctly — because a result computed on a contaminated specimen is worse than no result, and a lab that ships it once is not a lab anymore.

The read has exactly this discipline, and it is the sharpest line between an instrument and an opinion. It does not go into the company for weeks and verify every contract, every line of code, every customer review — that is a different business, a slow one, and refusing it is what keeps the read fast and neutral. The obligation to deliver correct data sits, squarely, with the leadership team. But no single number they deliver stands alone. Every figure is held against another — a claim against the reality it implies, one financial line against another it cannot contradict without breaking, the team’s own account against an operator’s outside eye, and each executive’s answer against the others’, given separately so no one can quietly align them. When one of those readings comes back implausible — a growth rate the company’s own cost base cannot support, a confidence that the numbers underneath it will not hold — the instrument surfaces it before any verdict is written. And then the same move a lab makes: go back to the team, name the contradiction, and require them to account for it before anything is certified.

That requery is the tell that separates this from the collapsing business next door. A model handed a set of numbers will produce a fluent analysis of them; it has no mechanism to know the numbers are lying, and no standing to make anyone answer for it. Surfacing that a company’s own account does not cohere, and stopping the verdict until it does, is not knowing. It is not doing either. It is the third thing the ladder has no rung for: the integrity of the reading itself.

There is a discipline that keeps this honest, and it is worth stating because it is where a lesser version of this would go wrong. Naming the contradiction is the lab’s job. Resolving it is not. “These two numbers cannot both be true — tell me which is wrong” protects the specimen. “Here is how to fix the number” is the surgery, and the moment a read starts prescribing the fix it has rebuilt the conflict that the neutrality was worth, and left the seat empty again. The discipline is to surface and hand back — never to resolve. A lab requeries the draw. It does not prescribe the diet.

One more part of the analogy carries weight: the patient. The company that benefits from a read is not the passive kind, waiting to be treated — it is the leadership team that wants the real diagnosis precisely because it intends to act on it. Not fear, not vanity: ownership. For a team like that, the honest reading is not a threat. It is the thing they were missing — a clear signal of what to fix, from something that cannot be argued with.

So the operator thesis is correct, and beside the point. Consulting sold knowing, knowing is collapsing, and the honest response is to go and do. But the read was never on that ladder. It draws a specimen nobody else collects, runs a panel nobody can dial, refuses to certify what it can see is corrupt, and hands a trusted result to the two parties who own the doing — the capital that decides, and the team that executes. The cheaper intelligence gets, the more that seat is worth — because when every opinion in the room is free, the only thing left with a price is the one result that cannot be argued with. The lab does not need to become the operator. It needs to keep refusing to.

Robert Kellner is co-founder of ClimeNow — Execution Intelligence for capital decisions in Energy & Resilience. Under Load reads what happens when capital meets a scaling company.

You know the position before I describe it, because you are already thinking about it. The one that isn’t failing — failing would at least be clear — but has quietly stopped moving. Revenue flat two quarters running. The founder calls shorter than they used to be. The milestone slide carries last quarter’s milestones under this quarter’s names. The board pack says on-plan, because mid-raise it has to. And it sits in the quiet middle of your portfolio review, where the positions go that no one has a sentence for.

The silence around that position feels like patience — like time you are generously extending while things sort themselves out — but time is not neutral to a stalled company in this sector. A hardware-and-software company that has stopped moving is consuming bridge capital while its strategic value decays, and there is a buyer who understands the decay perfectly: the acquirer running the salvage screen, the one party at the table with no clock at all, for whom every undiagnosed quarter is a discount. Your silence is their entry point. You experience it as waiting. They experience it as a sale slowly coming to them at a better price.

So the question that matters about the stall is not how much longer do we give it. It is what is actually holding it still — and this is where the quiet stall splits into two companies that look identical from the outside and could not be more different underneath.

The first company has a locatable, fixable cause. A production model built for a volume it never reached. A commercial architecture that worked at the pilot scale and shears at the real one. A seat in the leadership team that was adequate before the load and is failing under it now. These breaks are specific, and specific breaks can be cleared. For this company, a bridge with the fix attached is one of the best trades in the portfolio — modest capital, a repaired position, a restored mark, and an investor letter that reads like discipline rather than apology.

The second company looks the same from the outside and is not. Its cause is structural, and no intervention you can fund clears it. For this one, the bridge buys six more months of the same fracture, a deeper entry for the salvage buyer, and a mark you will eventually explain twice — once when it falls, and once when someone asks why fresh money went in ahead of the fall.

This is not hypothetical. Britishvolt stalled in public view through the autumn of 2022 — bridged twice, diagnosed never. January 2023: administration. Weeks later, its assets sold for a reported £8.6m, against a £3.8bn plan. The bridges bought months. Nobody had established what they were supposed to buy.

On the day you write the check, the two companies are indistinguishable. The board pack cannot separate them mid-raise; your conviction will not — conviction is what carried the position this far, and it argues, always, for one more quarter. So the extension gets made because deciding is hard and extending is easy, and the clock keeps running whether or not you have looked.

Without the read, there is no choice. There is only the clock.

That is the sentence I want to sit on, because it is exact. What feels like a decision — extend or exit, patience or discipline — stops being one the moment you cannot tell the two companies apart. It becomes a default. The clock advances, the runway shortens, and at the end of it something happens to you that you experience as something you chose. A choice requires two distinguishable options and a basis for picking between them. Strip out the basis and the options collapse into one, and the one that remains is simply: wait, and let the clock decide.

A read is what puts the choice back. Not a longer binder, not a second opinion, not more conviction applied to the same fog — a plain, reproducible reading of what is holding the position still, whether capital controls it, and whether a fix clears it. It does one thing to the quiet stall: it separates the fixable company from the structural one, before the extension, while intervention is still cheap. On the first, it hands you the best trade in the book and the evidence to defend it. On the second, it hands you the harder truth early, when early is worth something, instead of late, when the salvage buyer has already priced it.

And it changes the hardest conversation you have — the one across the table from the family office you are asking to co-fund the extension. Their clock and yours disagree by construction: your term pushes toward an exit they may not want, their patience allows a repair your fund life cannot wait for. Two clocks do not negotiate; they just run at each other. A read is the one object both of you can hold at once — the same reading, the same named blocker, for every holder at the table — which is the only honest basis on which a fund and a family office can fund an intervention together without one of them quietly underwriting the other’s timeline.

None of this stops the clock; nothing does. What the read decides is whether it runs while you watch, understanding exactly what you are holding and why — or while you wait, calling the wait a choice. The quiet stall will resolve either way. The only open question is whether it resolves as something you read and acted on, or something that happened to you at a price set by the one buyer who was never in a hurry.

Robert Kellner is co-founder of ClimeNow — Execution Intelligence for capital decisions in Energy & Resilience. Under Load reads what happens when capital meets a scaling company.

Two good things already exist at every capital table. Action plans get written, and often well — the 100-day plan, the value-creation roadmap. And capital gets gated — the Series B Part I, milestones in half the term sheets in this sector. Both halves of a disciplined deployment are on the table.

But look at what connects them: nothing — and not because anyone is failing at their job. The plan and the tranches were built by different parties for different purposes and coupling them was never anyone’s mandate. So, the plan stays advisory, worked or deferred as the quarters allow, and the tranches gate on what a term sheet can hold: revenue thresholds, bookings, signed contracts — negotiated numbers derived from no diagnosis.
A company can clear its revenue milestone dead-on while the design flaw underneath compounds, and the tranche releases into the same unread break. Gated capital, ungated causes.

What’s missing is a structure, not a service: a deployment program where the diagnosed blockers are the release conditions, evidence is the closing mechanism, and a re-read verifies. Not a replacement for the ops team, the advisor’s roadmap, or the fund’s tranche structure — the object that connects them.

So, this essay starts where the essay “Yes or No Is Not a Diagnosis” stopped: capital paused, fixes required first. What does the program between that verdict and the wire have to look like? Four requirements, each one closing a specific failure mode.

First requirement: the blockers are named and filed

A stall diagnosis worth acting on isn’t a mood about the company. It’s a finite list — these specific tensions block deployment; not fifty risks with no ranking, the pairs that bind, each traceable to the evidence that surfaced it.

And the localization from the essay named above is what turns the list into a schedule. Every tension has an address — design, current performance, or under load — and the address determines when it must clear.

• A tension in the design column means the plan wasn’t buildable as drawn; nothing built on top of it is real until it’s fixed.

• A tension in current performance means the machine is underdelivering today.

• A tension under load is the break waiting at exactly the volume the capital is buying.

• And a tension outside the company’s control belongs to no internal queue at all — it belongs on a watch list, with the external evidence named that would clear it.

The diagnosis doesn’t just find the breaks. It files them. And the filing is the schedule.

Second requirement: the gates are typed and ordered

Each blocker becomes a gate — the specific condition that must be true before it clears — and the addresses sort the gates into four kinds.

• External conditions stand watch above everything: the approval pending, the subsidy decision, closed only by outside evidence, and no capital should release against a company whose external ground is still moving.

• Then, inside the building: design fixes first — the tensions that mean the plan doesn’t hold as drawn.

• Operational fixes second — the machine brought to its stated performance.

• Scale preparation last — the under-load breaks resolved before the volume arrives that would trigger them.

The order runs through the work itself, for a reason that’s engineering, not process: scale preparation on top of an unfixed design is preparation for scaling the flaw. Within each stage, the items run in parallel — a stage is a front, worked across functions at once, not a queue. But the stages sequence each other, and the capital follows the same order, releasing on clearance, stage by stage.

A fixed order also closes a temptation no team is immune to — working the fast closers that show best while the deep design fix waits. Not the order of comfort or visibility: the order in which capital becomes safe, and nobody’s to renegotiate — either side of the table.

Third requirement: evidence closes, people don’t

What closes a gate is not a status update. Each gate needs its closing condition named in advance, on day one, verifiable by anyone: onboarding throughput at or above the installation rate, activation backlog at zero; runway extended past eighteen months at current burn. Specific enough that two people reading it reach the same answer — cleared, or not — with nothing left to argue.

Each gate also carries accountability, and the shape of it matters. A tension lives between two functions, so where it spans two executives, both are accountable together — single ownership of a two-function gap recreates the silo that caused it. And even a gate with one name on it is never one function’s work: a tension pair spans teams by nature — install crews and onboarding, engineering hours and field operations, finance and the production floor — so the resolution is a team play between functions every time, whoever’s name carries it.

And the clause that makes the whole structure survivable for a leadership team: the gate defines what must be true — never how. No imposed methodology, no outside party grading effort, and no OKR cascade — I say that from experience, not theory: cascade objectives down through an organization and a company that needed to fix five things ends up administering five hundred, with the coordination of the fragments consuming the capacity that was supposed to do the work.
A gate runs the opposite direction — few, held at leadership level, closed by evidence rather than reporting. Nothing cascades. The executives keep full freedom of method, and where they want help on the how, they bring in whoever they trust. The program names what must clear and hands it back. It does not prescribe the diet — a program that starts prescribing has become advice, a different seat at the table, and one that’s already well occupied.

Fourth requirement: the re-read decides

A quarter passes, or two, or three. Then the same questions get reviewed again, to the same standard, and the gates answer with evidence, not narrative. Cleared, or not cleared — and nobody argues about which.

This is where the verdict reveals what it was: a position on a path, not a stamp. The distinction carries real weight — a verdict delivered as a sentence does to a leadership team what it does to a patient: they stop fighting. Gates, owners, and named closing conditions are the opposite object: the shortest documented route back to deployable, and a team that can see the route fights for it.

A gate clears, the position lifts, a tranche releases — and the bridge written now is a different object from the bridge in the last essay: gated, evidenced, reaching a defined other side. When the gates don’t clear, that too is an answer, early enough to act on: the honest exit or the shutdown, on evidence, on the holder’s clock — not the salvage buyer’s.

Either way, the position on the evidence page stops being a silence. It becomes a schedule with a scoreboard.

The discussion: Pass is not a dead end — but it can be

Now the hardest diagnosis, because it’s the one everyone misreads. Pass can be terminal — some blockers are controlled by no one, and the honest end of that read is an orderly exit.

But Pass doesn’t have to be terminal, and the difference between the two is knowable. Pass means the company as constituted cannot clear its blockers. It says nothing yet about whether the holder can — because some blockers sit outside the executive team’s control entirely, and squarely inside the investor’s.

The leadership seat failing under load — the team can’t replace its own chief executive; the board can. The pivot the executives can see but cannot sanction — away from own-box manufacturing toward the ODM and service model, an identity change nobody inside wants to be the one to propose; the shareholders can propose it. The market that regulation just closed — the company can’t relocate its demand; the capital behind it can sanction the switch. In each case the diagnosis reads Pass on the company and hands the holder something better than a rejection: the named list of what would have to change, and who has the power to change it.

That three-way distinction — company-controllable, investor-controllable, controllable by no one — separates a Pass worth working from a Pass worth taking. A Pass whose blockers the holder controls isn’t a lost position; it’s a turnaround with a named lever. The alternative to knowing which kind you hold is the ending the last essay described: the undiagnosed drift, the bridge that was a pier, the salvage sale to the one buyer who was never in a hurry.

What patient capital actually buys

Put the four requirements in a row and the product of a diagnosis was never the verdict. It was the sequence: blockers named and filed, gates typed and ordered, closed by evidence, re-read on a clock — with fund, family office board, and leadership team holding the same object and the same scoreboard. The verdict is just the sequence’s first line.

That’s the difference between funding a conviction and funding a program. Fix first. Then protect the next check.

For a fund mid-raise, the same gate plan changes the story: a stalled position becomes a turnaround with evidence behind it, not a pier with fresh money on top. Better decisions on the way in are better capital efficiency on the way out — better DPI, better MOIC, fewer salvage exits.

For a family office the arithmetic is plainer still: the duration is already there, and what a gate plan protects is the willingness to use it. That is what makes this larger than one position. Every avoidable write-down pushes patient capital further out of Energy & Resilience, and that is one of the few sectors where long-duration money and industrial build-out actually have to meet. Keeping that capital in the room is the point — long enough to back the companies Europe needs built.

Robert Kellner is co-founder of ClimeNow — Execution Intelligence for capital decisions in Energy & Resilience. Under Load reads what happens when capital meets a scaling company.

Most capital decisions in this sector come down to a single word: yes or no. For a SaaS company, that compression is almost defensible — the asset is a subscription base, the failure mode is churn, a failed position gets marked down and left behind. For a heat pump company with a control platform, a field service organization, a supply chain through Shenzhen, and a revenue line gated by grid-connection approvals, yes or no is not a diagnosis. It’s a verdict in the absence of a diagnosis.

The question an IC actually needs answered: what is this capital deploying against — and what instruction follows?

Five instructions, one word

1. The company that’s fine, and needs nothing. Rare, and worth knowing precisely. Go·Deploy — full check, no gates.

2. The company that’s fine, against something. Go·Milestone — fund against named milestones, with named owners, and someone checks. Collapse this into the same “yes” as Go·Deploy and you’ve deployed unconditionally into a company that needed guardrails. Nobody notices for six quarters. Then everybody does.

3. The company blocked by something it controls. Install throughput at 50 a week against a plan that needs 120. Ten hours per install against a model priced at six hours. Controllable — so: Revisit·Execution — fix first, then fund, capital released against the sequence. Call it a no and you’ve walked away from an asset a competent operator turns in three quarters. Call it an unconditional yes and you’ve funded the stall.

4. The company blocked by something it doesn’t control. Grid queue. Subsidy regime under review. A capture price weakening as solar floods the day-ahead market. Structural and external — so: Revisit·External — hold, and name the external evidence that opens the gate. Treating this as a rejection is how European capital keeps walking past companies whose only sin was being early to a queue.

5. And the company where the blocks are structural and permanent. Pass — early, at a decision, where it’s cheap. The alternative is the bridge that simply extends the timeline without solving anything — and that bridge always gets written, because nobody could show the blocks were permanent.

Five situations, five verdicts: Go·Deploy, Go·Milestone, Revisit·Execution, Revisit·External, Pass. A yes/no vote carries exactly one of them — the other four get rounded off to whichever side of the line they fall on. That rounding error is where the money goes to die.

Where does the bridge go?

Everyone at these tables already knows the binary is wrong. The proof is the bridge.

A portfolio company stalls — revenue flat two quarters, the milestone slide carrying last quarter’s milestones under this quarter’s names — and watch what the holders do. They don’t vote yes: everyone at the table knows full deployment into a stalling company is crazy. They don’t vote no: nobody is prepared to let their own company die. So they invent a third state the binary doesn’t offer. That’s what a bridge is — an improvised verdict, built at the table. And because it’s improvised, it’s written ungated: money in, hope attached, no defined state it has to reach. The tensions that caused the stall are never named, so nothing clears between checks — and the same logic produces a second bridge, a third, a fourth. Serial bridging is the binary failing in slow motion, at €2–5M per installment.

A bridge is only a bridge if it reaches the other side — and “the company survives until the next board meeting” doesn’t qualify. The other side is a defined state: blockers cleared, the company back on track, the next round fundable. A Go·Milestone, not a wish. Write a bridge without knowing that state and you’ve built a pier. The company walks to the end of it with your money, and piers end the same two ways: a write-off, or a late exit at salvage price to a strategic acquirer buying the customer base and the service book for a fraction of the capital that built them.

Three questions come before the term sheet.

1. What is blocking the company — controllable or structural?

2. Does the money buy a sequence of fixes, or just time?

3. In what order do the blocks clear, and which two executives own each one?

With answers, the bridge gets written against gates and can work. Without them, the honest answer may be a pass — cheaper reached early, at a decision, than late, at a write-off.

A fund mid-raise structurally avoids these questions; a family office or balance-sheet investor, with no fundraise clock and no vintage optics, can afford them. Today they get answered with a feeling — usually “the team is strong and the market is real,” which is true of nearly every company that has ever burned €15M and stalled.

So: how do you keep a bridge from becoming a pier? You start by reading what the reporting cannot show.

The signal is between the numbers

Board packs, monthly reporting, four years of financials — and every number read in isolation, which is why the pack said on-plan right up until it didn’t.

A KPI on its own is nearly meaningless. The unit of diagnosis is the tension pair: two readings that must both be true, at the same time, for the plan to be real:

• Install speed against onboarding speed. Three days to put a system in the ground, eighteen to activate it. Both teams performing; the gap belongs to neither.

• Release speed against uptime. Six releases a month, 99.9% fleet uptime — both drawn from the same engineering hours. Every release is a small bet against the fleet.

• Customer satisfaction against growth rate. NPS at 40 in the installed base, collapsing in the last two cohorts. The blended number stays fine for three more quarters.

• Legacy fleet against new cohorts. Every unit shipped adds permanent service load. The base grows even in quarters sales doesn’t.

Each pair is a fact that exists only in the relation — and a dashboard is built, by construction, to show one number at a time. Neither the investor nor the CXO team can see it: each executive right about their own square, no one owning the distance between two functions. The board is shown an average, and an average is where dissent goes to die.

A pair, once found, still has to be located. Two questions do it.

1. Where does it live — in the design (the plan was never buildable as drawn), in current performance (the machine is underdelivering today), or under load (it holds now and breaks at exactly the volume the capital is buying)?

2. And whose domain is it — people, market and plan, or execution? Nine addresses, and the address decides the instruction: an install-throughput gap in current performance, execution domain, is Revisit·Execution. The same gap in design, market and plan, may be Pass.

The under-load column is the dangerous one — and the common mistake is treating it as unknowable, the part you only learn after the wire. It isn’t. Whether the under-load row holds is written in today’s pairs:

• revenue growth against cost degression,

• headcount against revenue per head,

• the certification clock against the cost curve.

Scheduling, logistics, and commissioning work fine at 50 installs a week and were never designed as one system — but whether they collapse at 1230 is not a mystery that arrives with the capital. It’s a reading available before it. The failure hasn’t happened yet; the evidence has. The company in the data room and the company after the wire only look like two different companies to whoever didn’t take the reading.

Then the question that decides the verdict: do the located tensions cluster? Scattered tensions are normal. Tensions that cluster into a pattern are a scale trap — a named, recurring way this sector kills companies:

• the Scaling Mirage (growth that exists in bookings and nowhere in the machine),

• the Regulatory Cliff (a demand assumption a ministry can delete),

• the Commoditization Cliff (below).

A trap has a state — forming, evolving, or already firing — and a trap that fires the day the wire lands was visible, in state and location, months earlier. That is what needs to be known before the check: the pair, its address, and the trap it feeds. Not after.

McKinsey named two valleys of death: research to prototype, and pilot to production. There’s a third: capital arrives, and the organization can’t absorb it. Nothing addresses it, because nobody measures whether the organization can carry the load before the load is applied.

The crossing is where Europe’s companies die

In this sector, that third valley is the graveyard, and the last three years put names on the stones. These are stacked companies — hardware, software, and a logistics-and-service model on different clocks, three failure surfaces where a software company has one — and they’re not flippable assets: a domestic storage manufacturer that dies takes a piece of Europe’s industrial capacity with it, and the replacement is imported.

Who got caught in the crossing — three kinds of company, one cause of death:

• The headline failures — Northvolt, Lilium, Sono Motors. None of them failed on the idea: each had raised, had product, had demand, and broke where capital met an organization that couldn’t carry the load it funded.

• The integrated platforms — the integrated platforms of the DACH market: hardware, energy management, installer networks, service, and financing in one model. Exactly the company type the transition needs most — and every interface between product, installation, service, and financing is a seam, and load finds seams. Not every strong brand is a load-bearing operating system.

• The installer and service wave — Eigensonne, Sungrade, Wegatech, and the distressed sales alongside: businesses that grew with the market and still died, because the machine underneath the growth couldn’t carry it in either direction.

And the next wave is already forming — the Commoditization Cliff, a scale trap in its purest form because it’s visible long before it fires. Chinese hardware now outprices and outperforms European product across cells, inverters, complete systems, while on the revenue side capture prices fall and negative-price hours multiply: every plain kWh earns less, and value migrates from producing energy to moving it.

The cliff itself is external and structural — nothing in a European building stops a cost and quality curve made in Shenzhen. But it announces itself in the unit margin, and what happens next is controllable: the survivors switch early from building the own-box manufacturing to ODM sourcing and a service-led model — owning the local customer relationship, the installation, the service book, the energy management, the financing — and let the Shenzhen curve work for them as falling input cost and rising input quality. That switch has a measurable window. Made in time, the cost curve becomes the margin; made too late, the company is competing against its own future supplier. What kills the companies in this wave is the late pivot — an execution call in plain sight that nobody was measuring.

Which reduces everything above to one question: what instrument lets a holder distinguish a fixable crossing problem from a structural one, before the check?

What patient capital needs

Withdrawal happens in steps, and every step is a measurement failure. A position stalls. Nothing measures it, so nobody knows what it is: a Revisit·Execution — fixable, in order, with owners — or a Pass that should have been called two quarters ago. Undetected, the tensions don’t get solved. Unsolved, the company drifts until the only buyer left is a strategic — increasingly a Chinese player — paying salvage price for the customer base, the service book, and the IP. The holder books a massive write-off, and the technology Europe funded scales under someone else’s flag. Run that sequence two or three times through a family office portfolio and the conclusion isn’t “tighten the process.” It’s “this category is uninvestable.” That salvage exit, repeated, is why patient capital leaves.

That’s the most expensive loss on the board: family offices, balance sheets, evergreen funds — the money that could hold a hardware company through its build years — walking away not because the sector can’t work, but because nobody could show them what they were deploying against, or that the stall in front of them was fixable. Patient capital can live without certainty. What it can’t live without is a read it can trust, so the difference between a fixable company and a broken one stops looking like luck.

That’s what I’m curious about — personally and professionally. What this sector needs is a diagnostic that runs the chain above end to end: clarity, transparency, and root cause, traceable from the verdict down to the tension pair, its location, and the scale trap it could feed. The read of that diagnostic decides nothing — the fund decides, the family office board decides, the company decides how to fix what gets named. What it establishes, reproducibly, is whether the break is controllable or structural, and in what order the blocks clear. The answer arrives as one of five verdicts a yes/no vote can’t carry: Go·Deploy, Go·Milestone, Revisit·Execution, Revisit·External, or Pass.

Not for a thesis. For the companies Europe needs to still be here in ten years — and for the people who’ll have to live with the grid they either did or didn’t build.

Robert Kellner is co-founder of ClimeNow — Execution Intelligence for capital decisions in Energy & Resilience. Under Load reads what happens when capital meets a scaling company.

There’s a conversation running through private equity about the death of the old return math: leverage costs too much, entry multiples have nowhere left to run, and the exit no longer arrives above the entry by default. Three of the four levers that built that industry were financial, and the one still standing — operational delivery — is the one its diligence machinery was never built to read.

Growth-stage Energy & Resilience should watch that conversation with a certain grim amusement, because it never had the financial levers to lose. No leverage to speak of, no multiple arbitrage, no engineered exit. A growth-stage company in this sector ever had exactly two levers: a market curve and a technology edge. The last three years took both. The market curve became unpredictable — demand moving on geopolitical anxiety and lobby outcomes, decoupled from electricity prices and payback math, at speeds no capacity plan absorbs. And the technology edge went onto the commoditization clock, where a lead built over a decade evaporates in a handful of quarters against a cost and quality curve made in Shenzhen.

What remains — for PE by subtraction, for growth-stage by construction — is the same lever: whether the organization can actually produce what the plan assumes. And here both worlds share a problem. The plan is fully instrumented. The organization is not.

Where the truth reports

Start with the number most plans are built on, because the problem begins there.

Charlie Munger’s view of EBITDA was that whenever you see the word, you should substitute “bullshit earnings” — and in this sector his point isn’t rhetoric, it’s engineering. Look at what the acronym deletes, letter by letter, in a hardware-and-software company. The D is the fleet: every unit shipped ages in the field, and its replacement and service burden is a real future bill arriving on a physical schedule — depreciation isn’t an accounting fiction here, it’s deferred cash. The A is the platform: the technical debt accumulating under every release the growth plan demanded. The I is the capital stack the whole asset-heavy model runs on. EBITDA removes precisely the three costs that kill stacked companies, and then presents what’s left as “operational performance.”

So a board pack demonstrating margin progress on EBITDA has demonstrated nothing about the company’s physics. The numbers where reality still reports are further down the page: EBIT at minimum, EBT honestly, and free cash flow always — the line that cannot be groomed, because the field crews, the suppliers, and the debt service all collect in cash. A growth-stage plan in this sector routinely demands 40–70% annual revenue growth, sustained across years, while the cost base holds. Whether that’s real is not visible in a metric designed to exclude the cost of the machine that has to produce it.

Two model types earn a partial defense, and naming them keeps the argument honest. In an ODM-sourced model, the company owns no fab — the hardware cost arrives as purchased COGS, inside EBITDA rather than deleted below it, and the D shrinks toward vehicles and tools. In a disciplined Hardware-as-a-Service structure, the fleet depreciation is still large, but it’s explicit, scheduled, and audited by whoever financed the fleet — the D isn’t hiding; it’s collateral. The critique lands with full force where it always did: the integrated own-manufacturing model — own product, own plant, own fleet, own service book — which is, not coincidentally, exactly the model the commoditization cliff is forcing out of this sector. The pivot these companies have to make anyway, from own-box manufacturing to ODM and service, also shrinks the room the metric has to hide things in. Until a company has made that pivot, read the lower lines.

Nobody goes to root cause

I’ve been on both sides of that table — sell-side and buy-side — and I’ll say it plainly: in none of those DD processes did anyone go to root cause. Not once. The only times I’ve seen a company’s actual stall mechanism get named were in board sessions, deep in the work, usually too late to be cheap. Never in a data room.

And it isn’t only the diligence that misses it. The management teams miss it too, in their own company, with all the data in front of them. End-to-end process reviews — lead to operations, order to cash, cash to upsell — get run silo by silo. Each function reports its own stretch, its own numbers, its own plan. The tensions between them, which is where the company actually breaks, stay untouched. Not hidden. Just nobody’s job.

The reason is rarely incompetence. It’s territory. “It’s business to fix this” — I’ve heard that from CSOs and COOs protecting authority in their own field. CFOs who won’t step into what they consider foreign territory: hardware manufacturing, customer service, technical debt. Every one of them competent. Every one of them looking at their own square.

What that looks like from inside

None of this shows up in a data room. All of it shows up the moment you put the same structured questions to every executive separately, run their answers against the financial model, and add an outside operator’s read of the same situation. Not because any single answer gives it away. Because the disagreements do.

The chairman wants the new modular IoT platform built greenfield. The CPO needs every person he has just to keep the existing platform standing under the growth already booked — he cannot staff a greenfield build, and he has not said so in those words. Both positions are defensible. Together they are a plan that cannot be executed, and nobody has put them in the same room as the same question.

The CSO reports a strong pipeline, and it is strong. He is also not supporting the COO’s end-to-end process build, because pipeline is his number and process is not. Both are performing well against their own scorecards. The company is losing the handoff between them.

The CFO can see the manufacturing partner will not ship on time. He sticks to the plan anyway, and gives the CPO’s ODM proposal less attention than it deserves. The reason isn’t financial. It’s cultural — shifting from being a German hardware manufacturer to being an ODM served by a Chinese global player is an identity change nobody wants to be the one to propose. So the risk stays in the model as a timing assumption rather than a decision anyone has made.

Four executives. Four rational positions. One plan that does not survive contact with all four at once.

That’s what an organization under load looks like before it fails. Not a bad number — a set of tensions nobody has surfaced, sitting exactly where the plan needs the most from the machine. And no leadership workshop surfaces them. No offsite, no trail in the Alps, no fifth whisky at the hotel bar. What comes out there is agreement — because that is what those settings are built to produce. The disagreements stay silent, and nobody in the room is counting what the silence costs.

The plan is not the organization

The distinction matters more than it sounds. A plan requiring 60% annual growth isn’t a financial object at all. It’s a claim about an organization — that throughput can rise while the cost base holds, that the leadership team agrees on where the growth actually comes from, that the operational machine underneath the model can produce what the model assumes. Diligence verifies the growth is in the model. It has no instrument for whether the organization exists that can deliver it.

And the claim runs in both directions, which almost nobody tests. A plan gets stress-tested against the upside case; the organization gets read for whether it can scale. Nobody asks the reverse question: if demand drops 30% for three quarters — and in this sector, it does — what in this cost structure is actually fixed, what is genuinely variable, and who has the authority to decide? That is not a forecast. It is a structural fact about the company, readable today. The insolvency wave of the last three years was, at bottom, a list of companies that had been built for the step-curve upward and only for that.

Diagnosis is not a forecast

Worth being precise here, because this is where most operational DD quietly overclaims.

No instrument can tell you whether a company will grow 60% a year. That’s a forecast, and forecasts about complex organizations are opinions with confidence intervals attached. Anyone selling certainty about the future of an operating business is selling you their judgment, dressed up.

What a diagnostic can tell you — deterministically, from evidence, the same way twice — is whether the conditions for that growth exist right now, and what specifically stands between the organization and the plan. Whether the executive team is aligned on the mechanism or only on the number. Whether the operational data supports the assumed unit economics or contradicts them. Which obstacles the company controls, and which are structural facts of its market that no amount of capital will move.

In the case above, that comes down to two questions with real answers: are the COO, CSO and CPO actually aligned on building the new modular platform — and has the ODM shift been decided, yes or no? Neither question is in the model. Both determine whether the model is worth anything.

That last distinction — controllable or structural — is what an investment committee actually needs. A company facing controllable obstacles is a rescue with a defined path. A company facing structural obstacles is a different asset entirely, whatever the model says. Writing the same check into both is how the last cycle’s portfolios got built — and those returns are public now: as of late 2025, only a quarter of 2021-vintage venture funds had returned any capital at all. The herd verified plans. Nobody read the organizations.

The question for the committee

Next time a plan crosses the table needing years of high double-digit growth to make the case, the question isn’t whether the model is right. The model is always right — about itself. And it isn’t whether the EBITDA bridge looks disciplined; that metric was designed to look disciplined.

The question is whether anyone has read the organization that has to produce the plan — in numbers that can’t be groomed, and in the disagreements the board pack was built to average away.

Robert Kellner is co-founder of ClimeNow — Execution Intelligence for capital decisions in Energy & Resilience. Under Load reads what happens when capital meets a scaling company.